#35 Internal coordinate singularity

[Joseph P. Kenny]

The torsions (including scaled torsions) created by
IntCoorGen have a singularity at 0 degrees, which
causes numerical problems during internal coordinate
optimizations with back-transformation to cartesian
coordinates. The warning message in this case is:

WARNING:
IntMolecularCoor::all_to_cartesian(RefSCVector&): too
many iterations in geometry update

I also suspect this to be the reason why bad
optimization steps may be generated when performing
cartesian optimizations using the internal coordinate
guess hessian. In the one observed case, this was
recognized by a strangely bad optimization step
followed by an erratic series of SCF iterations (due to
the bad geometry).

[Joseph P. Kenny]

Logged In: YES
user_id=1055622

Workaround: use cartesians for optimization and guess hessian

This problem becomes more common as the molecular size
increases.

[Joseph P. Kenny]

mpqc input file

[Joseph P. Kenny]

Logged In: YES
user_id=1055622

While admittedly a bit contrived, I am attaching an input
which I believe will trigger this condition. It certainly
fails, but I haven't revisited exactly why -- it's been a
while since I've dug into this stuff.